Abstract
Presently, service-based software systems (SBSSs) have been heavily deployed to fulfill the functionalities of cloud computing and widely used in many other application fields. The core technology of SBSSs, namely service-oriented computing (SOC), enables faster integration of existing software components (services) from different parties through standard protocols. By combining varied services, a system may be prototyped, built, and adapted quickly to fulfill the constant changing demands of the customers. SBSSs also make software fault tolerance (FT) feasible to provide continuous and reliable system service delivery when failures occur, by leveraging several functionally equivalent services in the repositories. In addition to the functional requirements, maintaining the quality of service (QoS) becomes increasingly important for the SBSSs, since failures for the systems to operate at expected service levels may cause great financial loss to an organization. However, the uncertainty nature of SBSSs, such as the server overloads and the network problems, actually introduces new challenges for system reliability evaluation. It is also noted that resource sharing commonly exists in different levels of the SBSS operations, such as the service providers, interconnected networks, or external libraries. Correlated failures due to the shared resources further complicate the SBSS reliability modeling, and they threaten the performance of existing FT designs. This dissertation presents a first study that considers the instability and dependency factors as well as the internal FT mechanisms in analyzing the reliability in SBSSs. We propose a reliability framework based upon the Markov process and varied theoretical tools. It incorporates the failure dependence modeling, the system reliability modeling, as well as the reliability analysis for the SBSSs with the FT designs. The proposed method is applied to a travel agency system based upon a real-world practice for verifying its accuracy of reliability modeling and effectiveness of varied reliability measures. In addition, we design a discrete-event SBSS simulation framework based upon the widely adopted ns-3 network simulator; it is capable of modeling the conditions of the underlying networks and could be also combined with various error models. Stochastic reliability and performance modeling for the enhanced SBSS FT schemes (ERB and ENVP) are also presented, discussed, and compared with the experimental results. Extensive results show that the proposed method could accurately model the SBSS reliability with varied workflow structures, FT designs, as well as the service and networking conditions. Various characteristics for the FT performance on SBSSs are uncovered. It is also confirmed that failure dependency of the services is an essential factor for analyzing any valuable SBSS. A set of reliability measures with different capabilities and complexities are available for assisting SBSS engineers with system improvements.